Professional Documents
Culture Documents
8 December 2016
Online at https://mpra.ub.uni-muenchen.de/75927/
MPRA Paper No. 75927, posted 02 Jan 2017 06:30 UTC
Strategy proofness and unanimity in many-to-one
matching markets
Mostapha Dissa , Ahmed Doghmi∗b , and Abdelmonaim Tlidic
a
University of Lyon, Lyon, F-69007, France; CNRS, GATE Lyon Saint-Etienne, Ecully,
F-69130, France; University of Jean Monnet, Saint-Etienne, F-42000, France.
b
University of Rabat, Mohammadia School of Engineering, the QSM Laboratory, Avenue Ibn
Sina B.P. 765 Agdal, 10100 Rabat, Morocco.
c
University of Marrakech, National School of Applied Science - Safi, Route Sidi Bouzid B.P.
63, 46000 Safi, Morocco.
December 8, 2016
1 Introduction
The matching problems are a branch of public decision problems that have been
extensively analyzed in the literature. They are two principle classes for these problems:
one-side matching markets and two-side matching markets. The first class studies the
assignment of a set of objects with finite capacities to some agents. As examples, we cite
the matching of office spaces to faculty members, the matching of student to dormitories
∗
Corresponding author.
Email addresses: diss@gate.cnrs.fr (M. Diss), ahmeddoghmi@hotmail.com (A. Doghmi), mtlidi2010@
gmail.com (A. Tlidi).
1
etc. The second class, that we are focusing on in this paper, is the one of the most
“popular” classes in matching theory; it explores two sides where each side is assumed to
have strict preferences over the opposite side. Gale and Shapley (1962) might be the first
to study these classes of matching theory. In these environments, stability is considered
as a key property to ensure a such assignment. It is checked if the two properties of
individual rationality and no-blocking are held. Individual rationality requires that each
agent finds her or his match acceptable. The property of no-blocking means that no
agent can improve by breaking up with their current match in order to match up with
each other instead. Gale and Shapley (1962) proved the existence of stable matchings by
providing an algorithm, called the Deferred Acceptance Algorithm, that has been a central
role to design matching programs across several institutions and markets. We distinguish
among three categories: one-to-one, many-to-one, and many-to-many matching problems.
The first category studies the assignment of each agent on one side on the market to at
most one agent on the other side. The best known example for this class is the problem
of marriages where there is a certain community with a set of men and a set of women,
where men only desire women and women only desire men that we would to match.
Each man ranks a set of women in accordance with his preferences for a mariage and vis
versa. For this class of matching, Gale and Shapley (1962) showed that there always is a
stable set of marriages. The second category establishes different examples of matching
problems in real world applications: on one side, colleges admit many student, firms hire
many workers, hospitals employ many interns, and on the other side, student attend one
college, workers work for one firm, and interns work for one hospital. The very known
example is the college admission problem that is introduced by Gale and Shapley (1962).
Each student ranks a set of college in accordance with his preferences for a matching
and each college ranks a set of students with a quota. They proved that a stable college
admissions matching exists. The third category analyses many situations where the very
known example is given by Roth and Sotomayor (1990) who examine the assignment
problem in the market for medical interns in the U.K. .
A solution is a logical process to select a set of matching outcomes. By solutions we
generally indicate matching rules which differ according some desirable properties. As
examples of matching rules: the stable rule, the Pareto rule, and the individually rational
rule.
Although the basic models of two side matching problems are based on the notion
of stability, many works in the matching literature are concerned with the question of
the strategic behavior of agents. Generally, in most cases of public decision problems,
agents have private information about their own preferences and they do not provide them
exactly. So that agents reveal truthfully their preferences, a central planner should find
mechanisms which give agents the incentive to reveal truthfully their private information
independently of the other agents’ behavior; the one of the most frequently mechanism
used in the recent literature in social choice theory is known as strategy-proofness.
This property is widely studied in the previous literature as long as it obligates the
institutions to reveal truthfully their preferences. It requires that no agent can manipulate
the outcomes of social choice rule in his favor by misrepresenting his preferences. It
constitutes a key property for the fundamental result in social choice theory, which is
known as the Gibbard (1973)-Satterthwaite (1975) Theorem. This result shows that if
there exist at least three alternatives and individual preferences are not restricted, then
2
every strategy-proof social choice rule is dictatorial.
In relation with matching problems, many authors studied the relation between
strategy-proofness and efficiency. In one-to-one matching markets, Roth (1982) provided
an impossibility result by showing that there is no stable and strategy-proof mechanism.
Alcade and Barberà (1994) proved that there is no solution that is Pareto efficient,
individually rational and strategy-proof. In many-to-one matching problems, Sönmez
(1996a) axiomatized matching rule and he proved that there is a rule that is Pareto
efficient, individually rational, and strategy-proof, which selects the unique stable
matching, if and only if the capacities are unlimited.
In this paper, we deal with many-to-one matching problems. As solutions, we focus
in particular on all sub-solutions of stable rule. Differently to the previous studies
that often relate the axiomatization of strategy-proofness with Pareto efficiency and
individual rationality1 , in this study we introduce the property of unanimity as a principle
requirement to masure efficiency. This intuitive property is very attractive and checked
by many solutions in the literature. We first provide a close connexion between strategy-
proofness and unanimity. Second, we use a part of the first connexion to determine the
relation between strategy-proofness and Maskin monotonicity. Alike connexion is given
recently but in a different topic by Diss et al. (2016) who explored the relation between
strategy-proofness, unanimity and Maskin monotonicity in a domain of single-peaked
preferences with private values. Their results provided a connexion with implementation
literature where these properties are central.
In our context and in relation with implementation theory, we show that there is a
close connexion with strategy-proofness and the implementability of the sub- solution
of the stable correspondence in dominant strategies equilibria and in Nash equilibria.
Before exposing our contribution in this direction, we start by giving a brief review of
implementation theory in matching problems. In these domains, many authors have
tried to apply a certain existing theoretical results to implement the stable SCCs in
various models. In the class of one-to-one matching models that include both marriage
problems (Gale and Shapley,1962) and housing markets (Shapley and Scarf, 1974),
Sönmez (1996b) used the results of Danilov (1992) and Yamato (1992) that are based
on a strong version of Maskin monotonicity. He proved that the solution of nonempty
core is Nash implementable with at least three agents. In one-to-one and many-to-
one matching problems, Kara and Sönmez (1996,1997) applied the results of Maskin
(1977,1999), Danilov (1992), and Yamato (1992). They examined the implementability
of certain examples of the stable matching SCCs. Haake and Klaus (2009) considered a
general class of many-to-one matching rules in studying a class of matching with contracts
markets (Hatfield and Milgrom, 2005). To extend the result of Tadenuma and Toda (1998)
from pure one-to-one matching rules to general many-to-one matching rules, Korpela
(2013) referred to a general variant of Maskin monotonicity developed in Korpela (2010)
and provided a full characterization for Nash implementation. In this topic, Doghmi and
Ziad (2015) have recently provided a full characterization based only on the property of
Maskin monotonicity.
Although several authors are interested to Nash implementability of stable rules in
matching problems, they are few works that examined the implementability of these rules
1
See for example Schummer (1997) and Hashimoto (2008) in the domain of exchange economies.
3
in dominant strategy equilibria. For instance, Kumano and Watabe (2012) studied the
implementability of the deferred acceptance algorithm, but on one-side matching markets,
where priorities objects cannot be private information and in using a direct mechanism.
They provided a full characterization based on strategy-proofness. In this paper we
study two-sided matching markets in which both preferences of workers and firms are
private information and prove that it is possible to implement the stable correspondence
in both dominant strategy equilibria and Nash equilibria by indirect mechanism. Hence,
we bridge the gap between Nash implementation and dominant strategy implementation.
We introduce a new condition and we prove that strategy-proofness is not only a necessary
property in general environments as known in the literature, but also becomes sufficient
for a sub-solution of the stable matching correspondence to be implementable with
at least three agents and at least three matchings when the requirement of citizen
sovereignty is fulfilled. In standard Nash implementation, strategy-proofness provides
a full characterization without any restriction on the number of matchings. However,
in partially honest Nash implementation, all sub-solutions of the stable matching
correspondences are implementable independently to strategy-proofness. From these
results we obtain a certain equivalence among these theories.
The rest of this paper is organized as follows: In Section 2, we introduce notations
and definitions. In Section 3, we state and prove our main result. In sections 4, we give
some implications. In Section 5, we study the relation between strategy-proofness and
implementation literature. In Section 6, we conclude.
4
A many-to-one matching µ is a mapping from the set F ∪ W into 2F ∪W such that:
For all w ∈ W, | µ(w) |≤ 1 and µ(w) ⊆ F;
For all f ∈ F, | µ(f ) |≤ qf and µ(f ) ⊆ W;
For all (f, w) ∈ F × W, µ(w) = f if and only if w ∈ µ(f ).
For any i ∈ F ∪ W, we call µ(i) the assignment of i under µ. We denote the set of all
possible matchings by M. We extend the preference relation Rf of a firm f ∈ F from
the class of subsets 2W∪{∅} to the set of matching M as follows: f prefers the matching
µ to the matching µ′ if and only if it prefers its assignment under µ to its assignment
under µ′ . Similarly, we extend the preference relation for a worker and we use Rf and
Rw to denote this extension.
A matching µ is blocked by a worker w ∈ W under a preference profile R if
∅Rw µ(w). A matching µ is blocked by a firm f ∈ F under a preference profile R if
µ(f ) 6= Hf (µ(f ); Rf ). Under responsive preferences, this statement is equivalent to the
following: A matching µ is blocked by a firm f ∈ F under a preference profile R if there
is a worker w ∈ µ(f ) such that ∅Rf w.
A matching µ is individually rational under a preference profile R if it is not blocked
by a worker or a firm under R; i.e., if µ(w)Pw ∅ for all w ∈ W and µ(f ) = Hf (µ(f ); Rf )
for all f ∈ F.
A matching µ is blocked by a firm-worker pair (f, w) ∈ F × W under a preference
profile R if (i) f Pw µ(w), and (ii) wPf w′ for some w′ ∈ µ(f ), or wPf ∅ if | µ(f ) |< qf .
A matching µ is stable under a preference profile R if it is both individually rational
and not blocked by any firm-worker pair.
We denote the set of all stable matchings under R by S(R). The stable many-to-one
matching rule is a correspondence ψ : L → M such that ψ(R) = S(R) for all R ∈ L . A
sub-solution of the stable many-to-one matching rule is any correspondence ϕ : L → M
such that ϕ(R) ⊆ S(R) for all R ∈ L . Gale and Shapley (1962) showed that a many-
to-one matching problem has at least one stable matching for any admissible preference
profile.
Now, we formally define the notion of strategy-proofness, which is the key property in
this paper. It requires that no agent can benefit from misrepresenting his true preferences.
Definition 1 (Strategy-proofness)
A sub-solution to the stable matching correspondence ϕ satisfies the strategy-proofness
property if for all R ∈ ℜ, all i ∈ F ∪ W, and all Ri′ ∈ ℜi , ϕ(R)Pi ϕ(Ri′ , R−i ).
The second fundamental property is unanimity. It means that if for all preference
profile, a matching is top-ranked for all agents, then this matching must be chosen.
Formally, it is defined as follows.
Definition 2 (Unanimity) The unanimity property states that if for all profile R ∈ ℜ,
for all i ∈ F ∪ W and for all matching ν ∈ M for which L(ν, Ri ) = M, then ν ∈ ϕ(R).
According to Doghmi and Ziad (2015), the property of unanimity is checked by all
sub-solution to the stable matching correspondence as shows the following proposition.
Proposition 1 (Doghmi and Ziad 2015) Any sub-solution to the stable matching
correspondence ϕ : L → M satisfies unanimity.
5
Next, we use the following condition.
This condition means that if in a first profile R, a socially chosen matching ν is strictly
dominated by a matching µ for an agent i , then there exists a new profile R, in which
the matching µ is chosen by the society, and ν improves her ranking and becomes top-
ranked for all agents. It has been recently introduced by Diss and al. (2016) to study
the relation between strategy proofness and unanimity in fair allocation problems with
single-peaked preferences. It is checked by all solutions satisfying the requirements of
citizen sovereignty 3 and external stability 4 in the problems of fair allocation.
In our context of many-to-one matching markets, we refer to the result of Roth (1982)5
and we provide the following example: let ϕ : L → M be a sub-solution to the stable
matching correspondence which associates the set of weakly Pareto efficient outcomes Pw
in M with a preference profile R defined as follows: ϕPw (R) = {µ ∈ M : ∄µ′ ∈ M such
that µ′ Pi µ for all i ∈ F ∪ W}.
We show that ϕPw satisfies Condition 1. Assume not, i.e., for ϕPw (R) ⊆ S(R) for all
R ∈ L , µ, ν ∈ M, i ∈ F ∪ W, and ν ∈ ϕPw (R), we have ν ∈ LS(µ, Ri ) (1), but for all
R ∈ L we have either (a) µ ∈ / ϕPw (R), or (b) L(ν, Rj ) 6= M for some j ∈ F ∪ W. For
statement (a), µ ∈ / ϕ (R) for all R ∈ L means that there exists µ′ ∈ M such that µ′ P i µ
Pw
for all i ∈ F ∪ W. Since statement (a) is verified for all R ∈ L , then we have for R, µ′ Pi µ
for all i ∈ F ∪ W. Hence, from (1) we obtain µ′ Pi ν for some i ∈ F ∪ W. If µ′ = ∅, then ν
can not be individually rational, and so it can not be stable, a contradiction. In all other
cases, there is a pair (f, w) ∈ F × W such that w ∈ µ′ (f ) and f ∈ µ′ (w), and we have:
(i) for some f ∈ µ′ (w), f Pw ν(w) and (ii) for some w ∈ µ′ (f ), wPf w′ for some w′ ∈ ν(f )
or wPf ∅ if | ν(f ) |< qf . Therefore, ν is blocked by a firm-worker pair (f, w) ∈ F × W
under R, and so it is not stable, a contradiction. For statement (b), we have for all
R ∈ L , L(ν, Rj ) 6= M for some j ∈ F ∪ W. This means that there exists µ′′ ∈ M such
that µ′′ Pi ν for some j ∈ F ∪ W, and hence we follow the same reasoning of statement (a).
Proof. Assume not. Therefore, there exists R and Ri′ such that ϕ(Ri′ , R−i )Pi ϕ(R).
Let ν ∈ ϕ(R) and µ ∈ ϕ(Ri′ , R−i ). Hence, ν ∈ LS(µ, Ri ) (1). By Condition 1, there
exists R ∈ L such that µ ∈ ϕ(R), and L(ν, Rj ) = M for all j ∈ F ∪ W (2). From
3
A sub-solution to the stable matching correspondence ϕ : L → M satisfies the property of citizen
sovereignty if for each µ ∈ M, there is a profile R ∈ L such that µ ∈ ϕ(R).
4
An SCF f (R) ⊆ X is externally stable under R if every allocation in X \ f (R) is dominated by f (R).
5
Roth (1982) proved that if a matching is an agent-optimal stable, then it is weakly Pareto optimal.
6
Proposition 1, ϕ satisfies unanimity, and hence ν ∈ ϕ(R). Therefore, ϕ(R) = {µ, ν}.
From (2), stability, and strictness we have µ = ν,6 which contradicts (1). Q.E.D
From Propositions 1 and 2 we give the following first main result in this paper.
The second property that has a close connexion with strategy-proofness is Maskin
monotonicity. It means that if a matching µ is socially chosen in a profile R and if the
matchings ranked below µ for all agents remain ranked below it (in a large sense) in a
new profile R′ , then the matching µ must be socially chosen in R′ .
7
have ωPi µ, and hence by (3), ωPi ν (4). By Condition 1, there exists R ∈ L such that
w ∈ ϕ(R), and L(ν, Rj ) = M for all j ∈ F ∪ W (5). By unanimity, ν ∈ ϕ(R), and hence
ϕ(R) = {w, ν}. From (5), stability, and strictness we have w = ν,8 which contradicts (4).
Q.E.D
From Propositions 2 and 3 we complete the proof of the second main result in this
paper.
Definition 4 (Y -monotonicity)
A sub-solution to the stable matching correspondence ϕ : L → M satisfies Y -
8
To prove this, we follow the same statement in Proposition 2.
8
monotonicity if for all R, R′ ∈ L , for any µ ∈ ϕ(R), if for any i ∈ F ∪ W,
L(µ, Ri ) \ {µ} ⊆ LS(µ, Ri′ ), then ϕ(R′ ) = {µ}.
This property requires that every possible ranking of matchings can be achieved from
an individual preference outcome.
From Corollary 1 of Yao and Yi (2007) and Proposition 5 in this paper we complete
the proof of the the following theorem.
9
Theorem 3 Let | M |≥ 3. If the property of citizen sovereignty holds, then a sub-
solution to the stable matching correspondence ϕ : L → M is implementable in dominant
strategies if and only if it satisfies Maskin monotonicity.
10
individual expresses her preferences in a lexicographic way. For a domain of single-peaked
preferences L , let Ci be the other components of the strategy space (which depends on
individual preferences, social states, . . . ). The set Si = L × Ci represents the strategy
profiles for a player i ∈ F ∪ W and S = S1 × ... × Sn is a set of strategy profiles. The
elements of S are denoted by s = (s1 , ..., sn ). A domain is a set D ⊂ L . For each
i ∈ F ∪ W, and R ∈ D, let τi (R) = {R} × Ci be the set of truthful messages of agent i.
We denote by si ∈ τi (R) a truthful strategy as player i is reporting the true preference
profile. We extend a player’s ordering over the set X to an ordering over strategy space
S. This is because, the players’ preference between being honest and dishonest depends
on strategies that the others played and the outcomes which they obtain. Let R i be the
preference of player i over S in preference profile R. The asymmetrical and symmetrical
parts of R R R
i are denoted respectively by ≻i and ∼i . Let Γ be a mechanism (game form)
represented by the pair (S, g), where Si = D × Ci and g : S → A is a payoff function.
Let N E(g, R , S) be the set of Nash equilibria of the game (Γ, R ). A mechanism
Γ = (S, g) implements a sub-solution to the stable matching correspondence ϕ : D → M
in Nash equilibria if for all R ∈ D, ϕ(R) = g(N E(g, R , S)). We say that ϕ is partially
honest implementable in Nash equilibria if there is a mechanism which implements it in
these equilibria. In this framework, we recall the Assumption A of Dutta and Sen (2012).
Assumption 1 There exists at least one partially honest individual and this fact is known
to the planner. However, the identity of this individual is not known to her.
Doghmi and Ziad (2012, 2013) showed that the properties of strict-weak no-veto power
and unanimity are sufficient for a social choice correspondence to be partially honest Nash
implementable.
Theorem 4 (Doghmi and Ziad (2012, 2013)) Let n ≥ 3 and suppose Assumption
A holds. If a sub-solution to the stable matching correspondence ϕ : D → M satisfies
strict-weak no-veto power and unanimity, then ϕ can be implemented in Nash equilibria.
According to Doghmi and Ziad (2015), the property of strict-weak no-veto power is
automatically checked for all sub-solution to the stable matching correspondence.
11
Proposition 5 (Doghmi and Ziad 2015) Any sub-solution to the stable matching
correspondence ϕ : D → M satisfies strict-weak no-veto power.
It follows from Theorem 3 and Propositions 1 and 6 that any sub-solution to the
stable matching correspondence can be implemented Nash equilibria in an environment
of partial honesty.
Corollary 5 Let n ≥ 3 and suppose Assumption A holds. Any sub-solution to the stable
matching correspondence ϕ : D → M is partially honest Nash implementable.
4 Conclusion
In this paper we have inspected the relation between strategy-proofness and unanimity.
We have first, introduced a new condition and we have proved that these properties
become equivalent if this new requirement is met. We have second, showed that
this result help us us to detect the link between strategy-proofness and Maskin
monotonicity. These results allowed us to determine a certain connexion between
strategy-proofness and implementation literature. In particular, we have proved that
dominant strategy implementation, standard Nash implementation, and partially honest
Nash implementation are equivalent under our new requirement.
References
[1] J. Alcalde and S. Barberà. Top dominance and the possibility of strategy-proof
stable solutions to matching problems. Economic Theory, 4:417–435, 1994.
12
[2] K.J Arrow. Social choice and individual values. 2nd Edition. Wiley, New York, 1963.
[3] D. Black. On the rationale of group decision making. The Journal of Political
Economy, 561:23–34, 1948.
[4] O. Bochet and T. Storcken. Maximal domains for Maskin monotone Pareto optimal
and anonymous rules. In: Collective Decision Making: Views from Social Choice
and Game Theory (Van Deemen, A. and Rusinowska, A. Eds), Springer, 43:57–68,
2010.
[6] M. Diss, A. Doghmi, and A. Tlidi. Strategy proofness and unanimity in private good
economies with single-peaked preferences. Working Paper, 2016.
[7] A. Doghmi. A simple necessary condition for partially honest Nash implementation.
Working paper, 2015.
[9] A. Doghmi and A. Ziad. On partially honest Nash implementation in private good
economies with restricted domains: A sufficient condition. The B.E.Journal of
Theoretical Economics, 13:1–14, 2013a.
[10] A. Doghmi and A. Ziad. Nash implementation in private good economies with single-
plateaued preferences and in matching problems. Mathematical Social Sciences,
73:32–39, 2015.
[11] B. Dutta and A. Sen. Nash implementation with partially honest individuals. Games
and Economic Behavior, 74:154–169, 2012.
[12] D. Gale and L. Shapley. College admission and the stability of marriage. American
Mathematical Monthly, 69:9–15, 1962.
[14] C.J Haake and B. Klaus. Monotonicity and Nash implementation in matching
markets with contracts. Economic Theory, 41:393–410, 2009.
13
[18] T. Kara and T. Sönmez. Nash implementation of matching rules. Journal of
Economic Theory, 68:425–439, 1996.
[20] N. Kartik, O. Tercieux, and R. Holden. Simple mechanisms and preferences for
honesty. Games and Economic Behavior, 83:284–290, 2014.
[21] B. Klaus and O. Bochet. The ralation between monotonicity and strategy-proofness.
Social Choice and Welfare, 40:41–63, 2013.
[26] E. Maskin. Nash equilibrium and welfare optimality. M.I.T. mimeo, 1977. Published
1999 in the Review of Economic Studies 66, 23-38.
[27] H. Mizukami and T. Wakayama. New necessary and sufficient conditions for secure
implementation. Working paper, 2016.
[30] Satterthwaite M.A. Muller, E. The equivalence of strong positive association and
strategy-proofness. Journal of Economic Theory, 14:412–418, 1977.
[31] J. Ortner. Direct implementation with minimal honest individuals. Games and
Economic Behavior, 90:1–16, 2015.
[32] A.E. Roth. The economics of matching: stability and incentives. Mathematics of
Operations Research, 7:617–628, 1982.
[33] A.E. Roth. The college admissions problem is not equivalent to the marriage problem.
Journal of Economic Theory, 36:277–288, 1985.
14
[35] T. Saijo, T. Sjöström, and T. Yamato. Secure implementation. Theoritical
Economics, 2:203–229, 2007.
[36] A. Saporiti. Securely implementable social choice rules with partially honest agents.
Journal of Economic Theory, 154:2016–228, 2014.
[44] H. Yao and J. Yi. Social choice rules implemented in dominant strategies. Economics
Letters, 97:197–200, 2007.
15